Robots that inspect what a camera can't.
Robots for the line and the field, returning one accept-or-reject call. Before yours ships, we build it with you in simulation — you watch it work on your parts first.
From the quality gate to the field.
Every robot runs the same models and decision layer, and carries the sensor head your parts demand — vision alone, or vision with phased-array ultrasound.
Bench cell
Parts come to the arm — fixtured, scanned, one call per part.
Inline cell
Over your conveyor, inspecting inside the takt — no parts pulled off the line.
Field crawler
On the weld itself — tanks, pipes, and hulls that can't come to a cell.
The costliest defects sit beneath the surface.
A camera reads the surface; the crack two millimetres down never reaches it. Our robots carry vision and phased-array ultrasound — thermography and eddy current to follow — and fuse every read into one accept-or-reject call.
- Surface
Camera (vision)
High-resolution imaging for surface cracks, geometry, and undercut. Necessary — but it only ever sees the outside.
- Surface / just below
Eddy current
Surface and just-below-surface cracks in metal — a fourth read that confirms or overrules the others.
- Near-surface
Thermography
Heat response reveals near-surface flaws and weak bonds, just under what the camera can see.
- Subsurface
Ultrasonic (PAUT)
Phased-array ultrasound sees inside the weld — cracks, lack of fusion, porosity no camera can reach. Our lead modality.
Any single sensor lets some defects through. Forcing several to agree on one call catches the flaw each of them, alone, would have missed.
A 99% accurate inspector can still bankrupt a quality budget.
Two inspectors can sort parts equally well on paper and cost you wildly different money — one lets defects escape to the field, the other just scraps good parts. So every call we make is priced: escape versus scrap, in dollars, against your acceptance class.
That's how you set the line — tighter where an escape is a recall, looser where a reject is a re-grind.
Same accuracy. 20× difference in what it costs to run.A percentage can't tell these two apart — dollars can.
Illustrative figures, to show the method, not a measured result.
Modeled, not measured — escaped-defect and false-reject savings only. Our full base case (50,000 parts/yr, 2% defects, $700/escape), with inspection labor and documentation counted, models ~$211k year-one benefit and 6.8-month payback. On a fit call we run the full model with your numbers.
Three kinds of part where a missed defect gets expensive.
We go deep on metal fabrication, where the costly defects are subsurface and one escape becomes a field failure.
Structural & heavy-fab welds
Porosity, lack of fusion, and cracks in energy, heavy-equipment, and industrial weldments — high-consequence, below the surface, and invisible to a camera.
Foundry castings
Surface and near-surface porosity, cracks, and inclusions checked against your ASTM or ISO acceptance class — without the fatigue and drift of all-manual inspection.
Forged & machined critical parts
Cracks and grinding burn in automotive-tier and industrial components, where one escape turns into a field failure downstream.
Different part? The 60-second fit check gives you a straight answer either way.
Your cell, proven before you commit.
You specify the part, we build your cell in simulation, and you watch it work before hardware ships. The same runs train the models that make the calls.
→ proven cell, verified on real parts
Bring us the defect that keeps escaping.
We're taking a small number of early partners in weld and casting fabrication. You work directly with the founders, on your parts, from the first conversation.
We read every reply ourselves.